jkarolczak / ligand-classification

Project examing sparse deep learning architectures for ligand classification.

Ligand classification using deep neural networks

Environment

Docker

We provide a docker setup with all dependencies and runtime configuration. To start the container follow:

1. Have installed:
   • docker (>=20.0.0)
   • docker compose (>=2.0.0!, notice, that the page suggests installing older version)
   • additionally, to run using GPU:
     • cuda (>=11.3) , nvcc and cudnn (>= 8.0)
     • nvidia docker runtime
2. Clone this repository.
3. sudo chmod 744 ./start.sh ./stop.sh
4. Set the environment configuration in the docker/.env:
   • PYTORCH, CUDA, CUDNN - only for GPU, modify only if needed. Before changing verify against Minkowski Engine requirements
   • DATA_PATH - if your name is Adaś and you are computer you should leave it as it is. If your name is Ania, Witek or Konrad, RNG guesses it is ../../data/. In other cases set the path to the folder containing data
5. Start container using ./start.sh. If you want to run container on cpu use ./start.sh cpu.
6. To stop use ./stop.sh or ./stop.sh cpu for cpu respectively.

PyCharm setup:

1. Mark src directory as the source root:
   • right click on src directory -> Mark Directory As -> Source root
   • with this step done, PyCharm will consider src as source directory, therefore to reach ligands-classification directory remember to use ../ in your code.
2. Exclude unnecessary directories from indexing:
   • PyCharm automatically indexes files in the project, so that it can propose i.e. path completion, yet some directories should not be indexed:
   • similarly to the first step: right click on directory -> Mark Directory As -> Excluded
   • directories to exclude: .neptune, data and venv if present
3. Branch management:
   • switch between the branches using PyCharm GUI, not with terminal tools
   • in the bottom-right corner, there is a special menu:

• select the desired branch and choose Checkout in order to switch branches,
• if you can't see a branch, make sure to run git fetch from the Git toolbar menu.

4. Pulling changes from main branch into a task branch:
   • should there be some update on the main branch, please make sure to include them and resolve potential merge conflicts before creating a pull request from your task branch
   • checkout to the main branch
   • update project using Git toolbar menu
   • checkout to the task branch
   • open branch menu -> choose main branch -> Merge 'main' into task branch
   • thanks to the above, all commits merged into the main branch will be applied to the specific branch

Data

Labels (.csv)

• All [download]
• Train set [download]
• Holdout [download]

Datasets

• Raw [download]
• Surface (neighbourhood=22) + uniform selection (max_blob_size=2000, method=basic) [download]
• Surface (neighbourhood=22) + uniform selection (max_blob_size=2000, method=max) [download]